Hybrid search: Why an AI search must also find article numbers
Image: AI-generated
A hybrid search combines an AI's search for meaning with the classic keyword search. It finds both circumscribed questions ("How long can I take vacation?") and exact terms such as article numbers, standards or names. For AI searches in companies, it is therefore usually the better choice than pure vector search.
A hybrid search combines an AI's search for meaning with the classic keyword search. It uses it to find both circumscribed questions ("How many days off am I entitled to?") and exact terms such as article numbers, standards or proper names. For AI searches and chatbots with company knowledge, it is therefore usually the better choice than a pure vector search.
Many AI searches impress in the demonstration and disappoint in everyday life: They understand colloquial questions, but cannot find the data sheet for "HX-2040". The reason lies in the way they search.
Two ways to search
Keyword search | Search for meaning (vectors) | |
|---|---|---|
Principle | finds documents with the same words | finds documents of similar importance |
Strong in | Article numbers, names, standards, abbreviations | Paraphrases, synonyms, whole questions |
Weak at | different choice of words | exact strings, rare technical terms |
Example | "DIN 18008" finds "DIN 18008" | "Holiday entitlement" finds "recreational leave" |
The meaning search uses an embedding model to convert texts into series of numbers that represent their meaning (more on this in the article Embeddings simply explained). However, an article number does not have a "meaning" that can be depicted well – which is why it often slips through in pure vector searches.
How hybrid search works
- The question is searched for twice: once by meaning, once by keywords.
- Both searches provide a ranking.
- The lists are merged – often using Reciprocal Rank Fusion (RRF), which combines by rankings instead of raw values.
- Optionally, another model re-evaluates the best hits (reranking).
- The best sections go to the language model, which formulates the answer from it.
The vector database Qdrant supports this directly: Several pre-searches with dense and thin vectors are merged using methods such as RRF or DBSF (according to Qdrant documentation, as of October 2026). Open WebUI also offers hybrid search for its knowledge collections.
Where hybrid search makes the difference
- Technical Support: Questions about device types and error codes
- Wholesale and production: article numbers, material designations
- Construction and Crafts: Standards, Items from Specifications
- Registries and administration: file numbers, paragraphs, names
- Care and Social Affairs: Abbreviations from Documentation Systems
What to check in an AI search
During a demonstration, ask not only nice questions, but also questions with article numbers, abbreviations and names from your everyday life. Check whether the answer names the source – this is the best protection against made-up answers, as the article AI answers with source citation explains.
Our AI document search uses hybrid search based on Qdrant to find questions in everyday language and exact terms in equal measure.
Counselling and accompanied implementation is eligible for funding: Through INQA coaching, the federal government covers 80% of the consulting costs of up to €1,200 per day for up to 12 days, i.e. up to €11,520 subsidy. In Mecklenburg-Western Pomerania and the other new federal states, BAFA consulting funding is also eligible at 80% until 31.12.2026 (50% in the old federal states). Funding is provided for consulting, not software. The funding check shows which program fits.
Frequently Asked Questions
What is hybrid search?
A search that combines meaning search (vectors) and keyword search and merges the results. In this way, both circumscribed questions and exact terms are found.
Why can't my AI search find item numbers?
Because pure vector search searches for meaning. Strings such as article numbers have hardly any meaning that an embedding model can map. An additional keyword search solves this.
What is Reciprocal Rank Fusion?
A procedure that merges several lists of results based on rankings. It also works if the search methods use very different rating scales.
Is hybrid search more expensive?
The additional effort is small: an additional index and a second query. Measured against better hits, it is almost always worthwhile.
Would you like to test how well an AI search finds your documents? Write to us via the contact page.
Muhamed Alahmed
With over 10 years’ experience in IT, I develop solutions that not only work from a technical perspective, but also create real added value and open up new possibilities.
More about bettersorted →More articles
Vector databases explained simply – using Qdrant as an example
What is a vector database, why do you need it for AI and what can Qdrant do? Clearly explained for decision-makers, with data protection tips. Read now.
Qdrant or pgvector? The right vector database for SMEs
Qdrant or pgvector for AI search in the enterprise? Differences in filters, hybrid search, operation and data volume – with clear decision-making support.
AI Answers with Source References: How to Avoid Hallucinations
Why AI hallucinates and how source references, RAG, clear rules, and tests prevent incorrect answers in the company – explained in a practical way. Read now.
Stay up to date on AI topics
Short updates on AI automation, funding programs and new posts — no spam, unsubscribe anytime.